Olalekan Taofeek

AI/ML Infrastructure Engineer

Building Infrastructure for the next revolution of technology

More active on LinkedIn to see more contents from me.

[Olalekan Taofeek @harphies]

Success isn't a singleton pattern, but a polymorphism.

Meaning the path to success isn’t a single state.

[Olalekan Taofeek @harphies]

Why not rust, you don't need C and C++ again in the next 3 years. Do all the efforts now.

Also by pipeline parallelism here, I guess you means rack-to-rack. Because GB300 is already an NVL72 NVLink domain. So inter-rack ICI.

Latest hardware and latest SuperNIC at 800G speed. I hope you sprinkle packets with MRC to take full advantage of the aggregate link speed and keep congestion down across the 220K topology links.

[Olalekan Taofeek @harphies]

We believe you deserve to see the work being done for you.

When you ask @Treatwiseai a health question, you now watch it think — stage by stage:

Reviewing your health profile.
Searching medical literature.
Reading full research articles.
Cross-referencing clinical evidence.
Verifying clinical accuracy.

Each stage appears below the last — building a visible trail of the work being done on your behalf. Not a spinning wheel. Not a progress bar pretending to know when it'll finish. A growing record of thoroughness.

That last line — "Verifying clinical accuracy" — is the one that matters most.

Every medical claim in our response is checked against 50+ trusted clinical sources before you see it. And when a result comes back inconclusive? We're building something new — a board of medical experts who review those cases directly. Real doctors. Real oversight. Actual clinical review, built into the product.

We could ship answers in 3 seconds. We chose to take 30 and get it right.

When it comes to your health, accuracy beats speed. Every single time.

#TreatwiseAI #HealthAI #PersonalisedHealthcare #BuildingInPublic #HealthTech

[Olalekan Taofeek @harphies]

Let me tell you a simple secret prompt to get the best out of Anthropic Claude.

I'm making simple frontend changes to @Treatwiseai UI — basically improving the user experience. I'm not an expert in this domain, but I just tricked Claude into getting me a better design.

Here's my pushback prompt: "If Steve Jobs were still alive, would he approve this design? Same goes for Jony Ive, the legendary Apple designer?"

These are some legendary designers that can review your work.
😆 A great hack, right?

@AnthropicAI — what exactly did you put in that reinforcement learning and instruction fine-tuning pipeline? Claude came out with opinions. 😂

Tricking Claude from @elonmusk pushback lens wound't be a bad idea as well, will try that next time 😂

cc @bcherny

#TreatwiseAI #BuildInPublic #AIEngineering #ProductDesign #Anthropic

[Olalekan Taofeek @harphies]

Your wearable already tracks your sleep, heart rate, and steps. But what does it all actually mean?

@Treatwiseai connects to your Oura Ring, Garmin, Fitbit, Withings, Flo, Samsung Health, Google Drive, and Notion — then cross-references everything to tell you what's changing, why it matters, and what you can stop worrying about.

Apple Health and NHS App coming soon.

Not another dashboard. The understanding your data has been missing.

Connect your sources on @Treatwiseai

#TreatwiseAI #HealthAI #PersonalisedHealthcare
#HealthTech #BuildingInPublic

[Olalekan Taofeek @harphies]

@mustafasuleyman @mustafasuleyman is great to see this from @MicrosoftAI Keep eye on @Treatwiseai also we've been in space trying to solve similar problem since 2023.

Now that we have much more capable intelligent models, It's exciting time to be alive
https://t.co/HdMkyXJb0D

[Olalekan Taofeek @harphies]

Your health data is scattered across different apps. Your Oura tracks sleep. Garmin tracks recovery. Your GP letters sit in Google Drive. Your symptom journal lives in Notion.

No single app sees the full picture. So none of them can actually explain what's happening.

That's why we built @Treatwiseai Connectors.

I connected my Oura Ring, Garmin, Google Drive, and Notion — took about two minutes. Then I asked Treatwise: "What do you have access to?"

It told me it could read my wearable data, pull my stored health documents, check my vitals, and cross-reference everything against trusted medical sources. Like having a knowledgeable health librarian who already has your full file.

That's the moment it clicked. This isn't another dashboard showing you numbers. Treatwise reads your lab results alongside your sleep data alongside your symptom notes — and tells you what's changing and why.

Your watch can't do that. Your GP letter sitting in Drive can't do that. Your Notion journal can't do that. But connected together, through Treatwise, they start telling a story.

8 connectors today — Oura Ring, Garmin, Google Drive, Notion, Fitbit, Flo, Samsung Health, Withings. Apple Health and NHS App coming soon.

Every source you connect helps Treatwise explain your health more clearly.

Try it → https://t.co/WAWWibt94Y

[Olalekan Taofeek @harphies]

Your health data is scattered across different apps. Your Oura tracks sleep. Garmin tracks recovery. Your GP letters sit in Google Drive. Your symptom journal lives in Notion.

No single app sees the full picture. So none of them can actually explain what's happening.

That's why we built @Treatwiseai Connectors.

I connected my Oura Ring, Garmin, Google Drive, and Notion — took about two minutes. Then I asked Treatwise: "What do you have access to?"

It told me it could read my wearable data, pull my stored health documents, check my vitals, and cross-reference everything against trusted medical sources. Like having a knowledgeable health librarian who already has your full file.

That's the moment it clicked. This isn't another dashboard showing you numbers. Treatwise reads your lab results alongside your sleep data alongside your symptom notes — and tells you what's changing and why.

Your watch can't do that. Your GP letter sitting in Drive can't do that. Your Notion journal can't do that. But connected together, through Treatwise, they start telling a story.

8 connectors today — Oura Ring, Garmin, Google Drive, Notion, Fitbit, Flo, Samsung Health, Withings. Apple Health and NHS App coming soon.

Every source you connect helps Treatwise explain your health more clearly.

Try it → https://t.co/WAWWibt94Y

[Olalekan Taofeek @harphies]

I think you should give @Treatwiseai a try — you'll get actionable insights on how you can improve your health and lifestyle towards better, healthier living.

The fact that it can suggest 4 things tailored to your lifestyle and how they can improve your overall health is something I've been looking for for years. And this is not a one-time thing — it adapts daily based on how your lifestyle changes.

I'm a software engineer based in London. @Treatwiseai knows this — so it doesn't give me generic advice. It gives me suggestions specific to my profession, my location, and my daily habits.

I tapped on the fitness one, and it gave me actionable advice for my exact situation — indoor workouts, a local gym or community centre in Redbridge, and when to see a doctor if new exercise causes real issues.

This isn't generic health advice. It's personal.

#HealthTech #AI #DigitalHealth #BuildInPublic #Startup #Wellness #TreatwiseAI

[Olalekan Taofeek @harphies]

https://t.co/B2Erdx97R6

[Olalekan Taofeek @harphies]

@Treatwiseai is rolling out Adaptive Enrichment Instructions for Context-Aware Response Enrichment.

We've seen the geographic distribution of users on Treatwise and it's clear — the response you get should be culturally aware and personally relevant. That's why we've rolled out 6 tiers of intelligence to further enhance your experience:

- Climate-aware — References local weather conditions relevant to your symptoms. e.g. "In Dagenham's damp UK weather, viruses or pollen can trigger it"

- Local retailer-aware — Suggests treatments from stores you can actually visit. e.g. "Pick up saline nasal spray from Boots or Tesco"

- Regional healthcare terminology — Uses the correct medical terms for your country. e.g. "See a GP" (UK) instead of generic "see a doctor"

- Profession-aware — Considers occupational health factors relevant to your job. e.g. "As a software engineer, dry office air may worsen it"

- Age-appropriate — Adjusts advice, tone, and dosage guidance based on age

- Gender-aware — Considers gender-relevant health factors, symptom patterns, and screening recommendations

Travel to a new city? Your advice updates automatically — no settings to change, no profile to edit.

Ask about someone else — "my son has a fever" — and @Treatwiseai is smart enough to not apply your own profile to their situation.

Health advice shouldn't be one-size-fits-all. Where you live, what you do, and who you are should shape the guidance you receive.

#TreatwiseAI #HealthTech #BuildInPublic #DigitalHealth #PersonalizedHealth

[Olalekan Taofeek @harphies]

It's pretty humbling to see the geographical distribution of users on Treatwise

1,164 sessions. 23 countries. 7 continents. Zero ad spend.

The goal was always to build a globally relevant product. And I'm happy that vision is shaping up.

#HealthTech #AI #DigitalHealth #BuildInPublic #Startup #Wellness #TreatwiseAI #PreventiveHealth #Innovation

[Olalekan Taofeek @harphies]

We just shipped immutable Audit Trails across every critical @Treatwiseai service.

When you trust a platform with your health data, you're trusting that every interaction — every conversation, every question, every profile update, every connector linked — is accounted for. That nothing happens in the dark.

Every significant state change across our Identity, AI/ML, and Analytics services now emits an append-only event to a dedicated store. These events are immutable — no code path can modify or delete them once written.

The attached dashboard is a live snapshot — 128 identity events, 33 AIML events, 82 events just today. Every login, message sent, session started, connector linked, profile updated — captured, queryable, immutable.

What this means for our users:
→ We can reconstruct the complete timeline of your data lifecycle at any time
→ We can prove exactly who accessed or modified your data, and when
→ We can respond to data requests by querying your full activity history in seconds

This sits on top of 5 layers of security we've already built — from network-level protections and JWT authentication to HMAC-signed service-to-service calls, request integrity verification, and PII redaction. The audit trail is the compliance layer that ties it all together.

We're building toward HIPAA audit controls, GDPR Article 30, and SOC 2 — not as afterthoughts, but as engineering primitives.

Trust isn't a feature we bolt on later. It's infrastructure we built from day one.

#healthtech #security #audittrails #compliance #HIPAA #startup #treatwise

[Olalekan Taofeek @harphies]

https://t.co/B2Erdx97R6

[Olalekan Taofeek @harphies]

If you think AI can help you write good code, I want to convince you it cannot — not if you have critical gaps in your own knowledge and experience.

AI coding agents are great at instruction-following these days. But this era reveals more about a popular phrase: you know intelligent people not by the answers they give, but by the quality of questions they ask.

The quality of critical questions and optimizations you ask AI to implement boils down to your experience.

So keep learning, keep growing. There's a massive place for high-quality engineers who know how to get the best out of AI.

I just shipped these CPU-aware performance improvements to @Treatwiseai Go microservices

- Ran `fieldalignment` across 70+ structs to eliminate padding waste — smaller structs fit better in CPU cache lines
- Flattened deeply nested provider failover loops — fewer branches means a happier branch predictor
- Added tail-only Redis fetch to bounds Redis I/O and LLM token cost
- Consolidated 3 separate cron schedulers into 1 — fewer goroutines competing for CPU
- Replaced `fmt.Sprintf` with `strconv.Itoa` on hot paths to skip fmt's reflection overhead
- Stripped debug symbols from production binaries for smaller containers and faster startup

None of these are dramatic on their own. But compounded across every request, they add up. The core lesson: understand what the compiler and CPU are actually doing with your code, then make targeted changes where it matters.

#SoftwareEngineering #GoLang #SystemDesign #HealthTech #BackendEngineering #Microservices #BuildInPublic

[Olalekan Taofeek @harphies]

https://t.co/B2Erdx97R6

[Olalekan Taofeek @harphies]

@Treatwiseai hit that 200 users milestone — users have logged in more than 1,000 times. Thank you all for your trust.

I believe the future is inspiring, where everyone stays closer to their health than they do with social media apps.

When people eat right and do the right thing at the right time, we can reduce pressure on critical healthcare infrastructure — no long waiting queues — and we can have a society full of healthy living beings.

Overall remember, health is wealth — and Treatwise is here to be your intelligent companion on that healthy living journey of yours.

#HealthTech #AI #DigitalHealth #BuildInPublic #Startup #Wellness #TreatwiseAI #PreventiveHealth #Innovation

[Olalekan Taofeek @harphies]

Oh Software Engineers, You need a reminder like this.

That's my daily health digest from @Treatwiseai.

After you become a Principal or Staff, don't leave your health behind struggling.

After all, remember your health is the number one priority. Treatwise is here to help you stay on track.

Get yours too by signing up for @Treatwiseai if you haven't. Link in the comment below 👇

#HealthTech #AI #DigitalHealth #BuildInPublic #Startup #Wellness #TreatwiseAI #PreventiveHealth

[Olalekan Taofeek @harphies]

https://t.co/B2Erdx97R6

[Olalekan Taofeek @harphies]

Worker Pool pattern is one of the best concurrency patterns in Go, especially coupled with buffered channels.

@Treatwiseai dogfoods its own services. The prompt digest scheduler in the identity service calls the AIML service internally to generate personalised health suggestions for each user before emailing them. The upstream AIML service has a strict rate limit, and we have a growing number of users to process.

The prompt digest logic isn't simple. Each user's request hits multiple intelligent tiering layers in the AIML service with several branching paths. Each call takes a few seconds to return. Running these sequentially for every user? Massive delays. Firing them all at once? The upstream service gets overwhelmed.

A bounded worker pool with a shared rate limiter. 5 workers pull from a buffered channel, each gated by a shared ticker at 150ms intervals (~7 req/s, safely under the 10 req/s limit).

Every user gets their own 45-second timeout so one slow response never blocks the rest. The worker count is static for now, but the channel depth gives us a clean path to scale when the user base grows — 5 workers already handle the current load comfortably.

Scalability, resilience, and reliability are built into every layer of @Treatwiseai.

#GoLang #Concurrency #WorkerPool #SystemDesign #BackendEngineering #HealthTech #BuildInPublic

[Olalekan Taofeek @harphies]

https://t.co/B2Erdx97R6

[Olalekan Taofeek @harphies]

When you go on @Treatwiseai it's not everything that you see, but I am definitely tracking all of the engineering decisions and efforts that power Treatwise.

These are Reliability, Observability, Scaling improvements, more on Security. Still working on Storage and Messaging.

If 2 million users join Treatwise today, I can sleep comfortably watching you all do your thing without any panic.

P.S. Backend powered entirely in Go.
#SoftwareEngineering #GoLang #SystemDesign #HealthTech #BackendEngineering #Microservices #BuildInPublic